2 * ring buffer based function tracer
4 * Copyright (C) 2007-2008 Steven Rostedt <srostedt@redhat.com>
5 * Copyright (C) 2008 Ingo Molnar <mingo@redhat.com>
7 * Originally taken from the RT patch by:
8 * Arnaldo Carvalho de Melo <acme@redhat.com>
10 * Based on code from the latency_tracer, that is:
11 * Copyright (C) 2004-2006 Ingo Molnar
12 * Copyright (C) 2004 William Lee Irwin III
14 #include <linux/ring_buffer.h>
15 #include <generated/utsrelease.h>
16 #include <linux/stacktrace.h>
17 #include <linux/writeback.h>
18 #include <linux/kallsyms.h>
19 #include <linux/seq_file.h>
20 #include <linux/smp_lock.h>
21 #include <linux/notifier.h>
22 #include <linux/irqflags.h>
23 #include <linux/debugfs.h>
24 #include <linux/pagemap.h>
25 #include <linux/hardirq.h>
26 #include <linux/linkage.h>
27 #include <linux/uaccess.h>
28 #include <linux/kprobes.h>
29 #include <linux/ftrace.h>
30 #include <linux/module.h>
31 #include <linux/percpu.h>
32 #include <linux/splice.h>
33 #include <linux/kdebug.h>
34 #include <linux/string.h>
35 #include <linux/rwsem.h>
36 #include <linux/ctype.h>
37 #include <linux/init.h>
38 #include <linux/poll.h>
39 #include <linux/gfp.h>
43 #include "trace_output.h"
45 #define TRACE_BUFFER_FLAGS (RB_FL_OVERWRITE)
48 * On boot up, the ring buffer is set to the minimum size, so that
49 * we do not waste memory on systems that are not using tracing.
51 int ring_buffer_expanded;
54 * We need to change this state when a selftest is running.
55 * A selftest will lurk into the ring-buffer to count the
56 * entries inserted during the selftest although some concurrent
57 * insertions into the ring-buffer such as trace_printk could occurred
58 * at the same time, giving false positive or negative results.
60 static bool __read_mostly tracing_selftest_running;
63 * If a tracer is running, we do not want to run SELFTEST.
65 bool __read_mostly tracing_selftest_disabled;
67 /* For tracers that don't implement custom flags */
68 static struct tracer_opt dummy_tracer_opt[] = {
72 static struct tracer_flags dummy_tracer_flags = {
74 .opts = dummy_tracer_opt
77 static int dummy_set_flag(u32 old_flags, u32 bit, int set)
83 * Kill all tracing for good (never come back).
84 * It is initialized to 1 but will turn to zero if the initialization
85 * of the tracer is successful. But that is the only place that sets
88 static int tracing_disabled = 1;
90 DEFINE_PER_CPU(int, ftrace_cpu_disabled);
92 static inline void ftrace_disable_cpu(void)
95 __this_cpu_inc(per_cpu_var(ftrace_cpu_disabled));
98 static inline void ftrace_enable_cpu(void)
100 __this_cpu_dec(per_cpu_var(ftrace_cpu_disabled));
104 static cpumask_var_t __read_mostly tracing_buffer_mask;
106 #define for_each_tracing_cpu(cpu) \
107 for_each_cpu(cpu, tracing_buffer_mask)
110 * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
112 * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
113 * is set, then ftrace_dump is called. This will output the contents
114 * of the ftrace buffers to the console. This is very useful for
115 * capturing traces that lead to crashes and outputing it to a
118 * It is default off, but you can enable it with either specifying
119 * "ftrace_dump_on_oops" in the kernel command line, or setting
120 * /proc/sys/kernel/ftrace_dump_on_oops to true.
122 int ftrace_dump_on_oops;
124 static int tracing_set_tracer(const char *buf);
126 #define MAX_TRACER_SIZE 100
127 static char bootup_tracer_buf[MAX_TRACER_SIZE] __initdata;
128 static char *default_bootup_tracer;
130 static int __init set_cmdline_ftrace(char *str)
132 strncpy(bootup_tracer_buf, str, MAX_TRACER_SIZE);
133 default_bootup_tracer = bootup_tracer_buf;
134 /* We are using ftrace early, expand it */
135 ring_buffer_expanded = 1;
138 __setup("ftrace=", set_cmdline_ftrace);
140 static int __init set_ftrace_dump_on_oops(char *str)
142 ftrace_dump_on_oops = 1;
145 __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
147 unsigned long long ns2usecs(cycle_t nsec)
155 * The global_trace is the descriptor that holds the tracing
156 * buffers for the live tracing. For each CPU, it contains
157 * a link list of pages that will store trace entries. The
158 * page descriptor of the pages in the memory is used to hold
159 * the link list by linking the lru item in the page descriptor
160 * to each of the pages in the buffer per CPU.
162 * For each active CPU there is a data field that holds the
163 * pages for the buffer for that CPU. Each CPU has the same number
164 * of pages allocated for its buffer.
166 static struct trace_array global_trace;
168 static DEFINE_PER_CPU(struct trace_array_cpu, global_trace_cpu);
170 int filter_current_check_discard(struct ring_buffer *buffer,
171 struct ftrace_event_call *call, void *rec,
172 struct ring_buffer_event *event)
174 return filter_check_discard(call, rec, buffer, event);
176 EXPORT_SYMBOL_GPL(filter_current_check_discard);
178 cycle_t ftrace_now(int cpu)
182 /* Early boot up does not have a buffer yet */
183 if (!global_trace.buffer)
184 return trace_clock_local();
186 ts = ring_buffer_time_stamp(global_trace.buffer, cpu);
187 ring_buffer_normalize_time_stamp(global_trace.buffer, cpu, &ts);
193 * The max_tr is used to snapshot the global_trace when a maximum
194 * latency is reached. Some tracers will use this to store a maximum
195 * trace while it continues examining live traces.
197 * The buffers for the max_tr are set up the same as the global_trace.
198 * When a snapshot is taken, the link list of the max_tr is swapped
199 * with the link list of the global_trace and the buffers are reset for
200 * the global_trace so the tracing can continue.
202 static struct trace_array max_tr;
204 static DEFINE_PER_CPU(struct trace_array_cpu, max_tr_data);
206 /* tracer_enabled is used to toggle activation of a tracer */
207 static int tracer_enabled = 1;
210 * tracing_is_enabled - return tracer_enabled status
212 * This function is used by other tracers to know the status
213 * of the tracer_enabled flag. Tracers may use this function
214 * to know if it should enable their features when starting
215 * up. See irqsoff tracer for an example (start_irqsoff_tracer).
217 int tracing_is_enabled(void)
219 return tracer_enabled;
223 * trace_buf_size is the size in bytes that is allocated
224 * for a buffer. Note, the number of bytes is always rounded
227 * This number is purposely set to a low number of 16384.
228 * If the dump on oops happens, it will be much appreciated
229 * to not have to wait for all that output. Anyway this can be
230 * boot time and run time configurable.
232 #define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */
234 static unsigned long trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
236 /* trace_types holds a link list of available tracers. */
237 static struct tracer *trace_types __read_mostly;
239 /* current_trace points to the tracer that is currently active */
240 static struct tracer *current_trace __read_mostly;
243 * trace_types_lock is used to protect the trace_types list.
245 static DEFINE_MUTEX(trace_types_lock);
248 * serialize the access of the ring buffer
250 * ring buffer serializes readers, but it is low level protection.
251 * The validity of the events (which returns by ring_buffer_peek() ..etc)
252 * are not protected by ring buffer.
254 * The content of events may become garbage if we allow other process consumes
255 * these events concurrently:
256 * A) the page of the consumed events may become a normal page
257 * (not reader page) in ring buffer, and this page will be rewrited
258 * by events producer.
259 * B) The page of the consumed events may become a page for splice_read,
260 * and this page will be returned to system.
262 * These primitives allow multi process access to different cpu ring buffer
265 * These primitives don't distinguish read-only and read-consume access.
266 * Multi read-only access are also serialized.
270 static DECLARE_RWSEM(all_cpu_access_lock);
271 static DEFINE_PER_CPU(struct mutex, cpu_access_lock);
273 static inline void trace_access_lock(int cpu)
275 if (cpu == TRACE_PIPE_ALL_CPU) {
276 /* gain it for accessing the whole ring buffer. */
277 down_write(&all_cpu_access_lock);
279 /* gain it for accessing a cpu ring buffer. */
281 /* Firstly block other trace_access_lock(TRACE_PIPE_ALL_CPU). */
282 down_read(&all_cpu_access_lock);
284 /* Secondly block other access to this @cpu ring buffer. */
285 mutex_lock(&per_cpu(cpu_access_lock, cpu));
289 static inline void trace_access_unlock(int cpu)
291 if (cpu == TRACE_PIPE_ALL_CPU) {
292 up_write(&all_cpu_access_lock);
294 mutex_unlock(&per_cpu(cpu_access_lock, cpu));
295 up_read(&all_cpu_access_lock);
299 static inline void trace_access_lock_init(void)
303 for_each_possible_cpu(cpu)
304 mutex_init(&per_cpu(cpu_access_lock, cpu));
309 static DEFINE_MUTEX(access_lock);
311 static inline void trace_access_lock(int cpu)
314 mutex_lock(&access_lock);
317 static inline void trace_access_unlock(int cpu)
320 mutex_unlock(&access_lock);
323 static inline void trace_access_lock_init(void)
329 /* trace_wait is a waitqueue for tasks blocked on trace_poll */
330 static DECLARE_WAIT_QUEUE_HEAD(trace_wait);
332 /* trace_flags holds trace_options default values */
333 unsigned long trace_flags = TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK |
334 TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO | TRACE_ITER_SLEEP_TIME |
335 TRACE_ITER_GRAPH_TIME;
337 static int trace_stop_count;
338 static DEFINE_SPINLOCK(tracing_start_lock);
341 * trace_wake_up - wake up tasks waiting for trace input
343 * Simply wakes up any task that is blocked on the trace_wait
344 * queue. These is used with trace_poll for tasks polling the trace.
346 void trace_wake_up(void)
350 if (trace_flags & TRACE_ITER_BLOCK)
353 * The runqueue_is_locked() can fail, but this is the best we
357 if (!runqueue_is_locked(cpu))
358 wake_up(&trace_wait);
362 static int __init set_buf_size(char *str)
364 unsigned long buf_size;
368 buf_size = memparse(str, &str);
369 /* nr_entries can not be zero */
372 trace_buf_size = buf_size;
375 __setup("trace_buf_size=", set_buf_size);
377 static int __init set_tracing_thresh(char *str)
379 unsigned long threshhold;
384 ret = strict_strtoul(str, 0, &threshhold);
387 tracing_thresh = threshhold * 1000;
390 __setup("tracing_thresh=", set_tracing_thresh);
392 unsigned long nsecs_to_usecs(unsigned long nsecs)
397 /* These must match the bit postions in trace_iterator_flags */
398 static const char *trace_options[] = {
426 { trace_clock_local, "local" },
427 { trace_clock_global, "global" },
433 * trace_parser_get_init - gets the buffer for trace parser
435 int trace_parser_get_init(struct trace_parser *parser, int size)
437 memset(parser, 0, sizeof(*parser));
439 parser->buffer = kmalloc(size, GFP_KERNEL);
448 * trace_parser_put - frees the buffer for trace parser
450 void trace_parser_put(struct trace_parser *parser)
452 kfree(parser->buffer);
456 * trace_get_user - reads the user input string separated by space
457 * (matched by isspace(ch))
459 * For each string found the 'struct trace_parser' is updated,
460 * and the function returns.
462 * Returns number of bytes read.
464 * See kernel/trace/trace.h for 'struct trace_parser' details.
466 int trace_get_user(struct trace_parser *parser, const char __user *ubuf,
467 size_t cnt, loff_t *ppos)
474 trace_parser_clear(parser);
476 ret = get_user(ch, ubuf++);
484 * The parser is not finished with the last write,
485 * continue reading the user input without skipping spaces.
488 /* skip white space */
489 while (cnt && isspace(ch)) {
490 ret = get_user(ch, ubuf++);
497 /* only spaces were written */
507 /* read the non-space input */
508 while (cnt && !isspace(ch)) {
509 if (parser->idx < parser->size - 1)
510 parser->buffer[parser->idx++] = ch;
515 ret = get_user(ch, ubuf++);
522 /* We either got finished input or we have to wait for another call. */
524 parser->buffer[parser->idx] = 0;
525 parser->cont = false;
528 parser->buffer[parser->idx++] = ch;
538 ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt)
546 if (s->len <= s->readpos)
549 len = s->len - s->readpos;
552 ret = copy_to_user(ubuf, s->buffer + s->readpos, cnt);
562 static ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
567 if (s->len <= s->readpos)
570 len = s->len - s->readpos;
573 ret = memcpy(buf, s->buffer + s->readpos, cnt);
582 * ftrace_max_lock is used to protect the swapping of buffers
583 * when taking a max snapshot. The buffers themselves are
584 * protected by per_cpu spinlocks. But the action of the swap
585 * needs its own lock.
587 * This is defined as a arch_spinlock_t in order to help
588 * with performance when lockdep debugging is enabled.
590 * It is also used in other places outside the update_max_tr
591 * so it needs to be defined outside of the
592 * CONFIG_TRACER_MAX_TRACE.
594 static arch_spinlock_t ftrace_max_lock =
595 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
597 unsigned long __read_mostly tracing_thresh;
599 #ifdef CONFIG_TRACER_MAX_TRACE
600 unsigned long __read_mostly tracing_max_latency;
603 * Copy the new maximum trace into the separate maximum-trace
604 * structure. (this way the maximum trace is permanently saved,
605 * for later retrieval via /sys/kernel/debug/tracing/latency_trace)
608 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
610 struct trace_array_cpu *data = tr->data[cpu];
611 struct trace_array_cpu *max_data;
614 max_tr.time_start = data->preempt_timestamp;
616 max_data = max_tr.data[cpu];
617 max_data->saved_latency = tracing_max_latency;
618 max_data->critical_start = data->critical_start;
619 max_data->critical_end = data->critical_end;
621 memcpy(max_data->comm, tsk->comm, TASK_COMM_LEN);
622 max_data->pid = tsk->pid;
623 max_data->uid = task_uid(tsk);
624 max_data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
625 max_data->policy = tsk->policy;
626 max_data->rt_priority = tsk->rt_priority;
628 /* record this tasks comm */
629 tracing_record_cmdline(tsk);
633 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
635 * @tsk: the task with the latency
636 * @cpu: The cpu that initiated the trace.
638 * Flip the buffers between the @tr and the max_tr and record information
639 * about which task was the cause of this latency.
642 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
644 struct ring_buffer *buf = tr->buffer;
646 if (trace_stop_count)
649 WARN_ON_ONCE(!irqs_disabled());
650 arch_spin_lock(&ftrace_max_lock);
652 tr->buffer = max_tr.buffer;
655 __update_max_tr(tr, tsk, cpu);
656 arch_spin_unlock(&ftrace_max_lock);
660 * update_max_tr_single - only copy one trace over, and reset the rest
662 * @tsk - task with the latency
663 * @cpu - the cpu of the buffer to copy.
665 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
668 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
672 if (trace_stop_count)
675 WARN_ON_ONCE(!irqs_disabled());
676 arch_spin_lock(&ftrace_max_lock);
678 ftrace_disable_cpu();
680 ret = ring_buffer_swap_cpu(max_tr.buffer, tr->buffer, cpu);
684 * We failed to swap the buffer due to a commit taking
685 * place on this CPU. We fail to record, but we reset
686 * the max trace buffer (no one writes directly to it)
687 * and flag that it failed.
689 trace_array_printk(&max_tr, _THIS_IP_,
690 "Failed to swap buffers due to commit in progress\n");
695 WARN_ON_ONCE(ret && ret != -EAGAIN && ret != -EBUSY);
697 __update_max_tr(tr, tsk, cpu);
698 arch_spin_unlock(&ftrace_max_lock);
700 #endif /* CONFIG_TRACER_MAX_TRACE */
703 * register_tracer - register a tracer with the ftrace system.
704 * @type - the plugin for the tracer
706 * Register a new plugin tracer.
708 int register_tracer(struct tracer *type)
709 __releases(kernel_lock)
710 __acquires(kernel_lock)
716 pr_info("Tracer must have a name\n");
720 if (strlen(type->name) > MAX_TRACER_SIZE) {
721 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE);
726 * When this gets called we hold the BKL which means that
727 * preemption is disabled. Various trace selftests however
728 * need to disable and enable preemption for successful tests.
729 * So we drop the BKL here and grab it after the tests again.
732 mutex_lock(&trace_types_lock);
734 tracing_selftest_running = true;
736 for (t = trace_types; t; t = t->next) {
737 if (strcmp(type->name, t->name) == 0) {
739 pr_info("Tracer %s already registered\n",
747 type->set_flag = &dummy_set_flag;
749 type->flags = &dummy_tracer_flags;
751 if (!type->flags->opts)
752 type->flags->opts = dummy_tracer_opt;
753 if (!type->wait_pipe)
754 type->wait_pipe = default_wait_pipe;
757 #ifdef CONFIG_FTRACE_STARTUP_TEST
758 if (type->selftest && !tracing_selftest_disabled) {
759 struct tracer *saved_tracer = current_trace;
760 struct trace_array *tr = &global_trace;
763 * Run a selftest on this tracer.
764 * Here we reset the trace buffer, and set the current
765 * tracer to be this tracer. The tracer can then run some
766 * internal tracing to verify that everything is in order.
767 * If we fail, we do not register this tracer.
769 tracing_reset_online_cpus(tr);
771 current_trace = type;
772 /* the test is responsible for initializing and enabling */
773 pr_info("Testing tracer %s: ", type->name);
774 ret = type->selftest(type, tr);
775 /* the test is responsible for resetting too */
776 current_trace = saved_tracer;
778 printk(KERN_CONT "FAILED!\n");
781 /* Only reset on passing, to avoid touching corrupted buffers */
782 tracing_reset_online_cpus(tr);
784 printk(KERN_CONT "PASSED\n");
788 type->next = trace_types;
792 tracing_selftest_running = false;
793 mutex_unlock(&trace_types_lock);
795 if (ret || !default_bootup_tracer)
798 if (strncmp(default_bootup_tracer, type->name, MAX_TRACER_SIZE))
801 printk(KERN_INFO "Starting tracer '%s'\n", type->name);
802 /* Do we want this tracer to start on bootup? */
803 tracing_set_tracer(type->name);
804 default_bootup_tracer = NULL;
805 /* disable other selftests, since this will break it. */
806 tracing_selftest_disabled = 1;
807 #ifdef CONFIG_FTRACE_STARTUP_TEST
808 printk(KERN_INFO "Disabling FTRACE selftests due to running tracer '%s'\n",
817 void unregister_tracer(struct tracer *type)
821 mutex_lock(&trace_types_lock);
822 for (t = &trace_types; *t; t = &(*t)->next) {
826 pr_info("Tracer %s not registered\n", type->name);
832 if (type == current_trace && tracer_enabled) {
835 if (current_trace->stop)
836 current_trace->stop(&global_trace);
837 current_trace = &nop_trace;
840 mutex_unlock(&trace_types_lock);
843 static void __tracing_reset(struct ring_buffer *buffer, int cpu)
845 ftrace_disable_cpu();
846 ring_buffer_reset_cpu(buffer, cpu);
850 void tracing_reset(struct trace_array *tr, int cpu)
852 struct ring_buffer *buffer = tr->buffer;
854 ring_buffer_record_disable(buffer);
856 /* Make sure all commits have finished */
858 __tracing_reset(buffer, cpu);
860 ring_buffer_record_enable(buffer);
863 void tracing_reset_online_cpus(struct trace_array *tr)
865 struct ring_buffer *buffer = tr->buffer;
868 ring_buffer_record_disable(buffer);
870 /* Make sure all commits have finished */
873 tr->time_start = ftrace_now(tr->cpu);
875 for_each_online_cpu(cpu)
876 __tracing_reset(buffer, cpu);
878 ring_buffer_record_enable(buffer);
881 void tracing_reset_current(int cpu)
883 tracing_reset(&global_trace, cpu);
886 void tracing_reset_current_online_cpus(void)
888 tracing_reset_online_cpus(&global_trace);
891 #define SAVED_CMDLINES 128
892 #define NO_CMDLINE_MAP UINT_MAX
893 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
894 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
895 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
896 static int cmdline_idx;
897 static arch_spinlock_t trace_cmdline_lock = __ARCH_SPIN_LOCK_UNLOCKED;
899 /* temporary disable recording */
900 static atomic_t trace_record_cmdline_disabled __read_mostly;
902 static void trace_init_cmdlines(void)
904 memset(&map_pid_to_cmdline, NO_CMDLINE_MAP, sizeof(map_pid_to_cmdline));
905 memset(&map_cmdline_to_pid, NO_CMDLINE_MAP, sizeof(map_cmdline_to_pid));
909 int is_tracing_stopped(void)
911 return trace_stop_count;
915 * ftrace_off_permanent - disable all ftrace code permanently
917 * This should only be called when a serious anomally has
918 * been detected. This will turn off the function tracing,
919 * ring buffers, and other tracing utilites. It takes no
920 * locks and can be called from any context.
922 void ftrace_off_permanent(void)
924 tracing_disabled = 1;
926 tracing_off_permanent();
930 * tracing_start - quick start of the tracer
932 * If tracing is enabled but was stopped by tracing_stop,
933 * this will start the tracer back up.
935 void tracing_start(void)
937 struct ring_buffer *buffer;
940 if (tracing_disabled)
943 spin_lock_irqsave(&tracing_start_lock, flags);
944 if (--trace_stop_count) {
945 if (trace_stop_count < 0) {
946 /* Someone screwed up their debugging */
948 trace_stop_count = 0;
953 /* Prevent the buffers from switching */
954 arch_spin_lock(&ftrace_max_lock);
956 buffer = global_trace.buffer;
958 ring_buffer_record_enable(buffer);
960 buffer = max_tr.buffer;
962 ring_buffer_record_enable(buffer);
964 arch_spin_unlock(&ftrace_max_lock);
968 spin_unlock_irqrestore(&tracing_start_lock, flags);
972 * tracing_stop - quick stop of the tracer
974 * Light weight way to stop tracing. Use in conjunction with
977 void tracing_stop(void)
979 struct ring_buffer *buffer;
983 spin_lock_irqsave(&tracing_start_lock, flags);
984 if (trace_stop_count++)
987 /* Prevent the buffers from switching */
988 arch_spin_lock(&ftrace_max_lock);
990 buffer = global_trace.buffer;
992 ring_buffer_record_disable(buffer);
994 buffer = max_tr.buffer;
996 ring_buffer_record_disable(buffer);
998 arch_spin_unlock(&ftrace_max_lock);
1001 spin_unlock_irqrestore(&tracing_start_lock, flags);
1004 void trace_stop_cmdline_recording(void);
1006 static void trace_save_cmdline(struct task_struct *tsk)
1010 if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
1014 * It's not the end of the world if we don't get
1015 * the lock, but we also don't want to spin
1016 * nor do we want to disable interrupts,
1017 * so if we miss here, then better luck next time.
1019 if (!arch_spin_trylock(&trace_cmdline_lock))
1022 idx = map_pid_to_cmdline[tsk->pid];
1023 if (idx == NO_CMDLINE_MAP) {
1024 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
1027 * Check whether the cmdline buffer at idx has a pid
1028 * mapped. We are going to overwrite that entry so we
1029 * need to clear the map_pid_to_cmdline. Otherwise we
1030 * would read the new comm for the old pid.
1032 pid = map_cmdline_to_pid[idx];
1033 if (pid != NO_CMDLINE_MAP)
1034 map_pid_to_cmdline[pid] = NO_CMDLINE_MAP;
1036 map_cmdline_to_pid[idx] = tsk->pid;
1037 map_pid_to_cmdline[tsk->pid] = idx;
1042 memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
1044 arch_spin_unlock(&trace_cmdline_lock);
1047 void trace_find_cmdline(int pid, char comm[])
1052 strcpy(comm, "<idle>");
1056 if (WARN_ON_ONCE(pid < 0)) {
1057 strcpy(comm, "<XXX>");
1061 if (pid > PID_MAX_DEFAULT) {
1062 strcpy(comm, "<...>");
1067 arch_spin_lock(&trace_cmdline_lock);
1068 map = map_pid_to_cmdline[pid];
1069 if (map != NO_CMDLINE_MAP)
1070 strcpy(comm, saved_cmdlines[map]);
1072 strcpy(comm, "<...>");
1074 arch_spin_unlock(&trace_cmdline_lock);
1078 void tracing_record_cmdline(struct task_struct *tsk)
1080 if (atomic_read(&trace_record_cmdline_disabled) || !tracer_enabled ||
1084 trace_save_cmdline(tsk);
1088 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
1091 struct task_struct *tsk = current;
1093 entry->preempt_count = pc & 0xff;
1094 entry->pid = (tsk) ? tsk->pid : 0;
1095 entry->lock_depth = (tsk) ? tsk->lock_depth : 0;
1097 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
1098 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
1100 TRACE_FLAG_IRQS_NOSUPPORT |
1102 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
1103 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
1104 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
1106 EXPORT_SYMBOL_GPL(tracing_generic_entry_update);
1108 struct ring_buffer_event *
1109 trace_buffer_lock_reserve(struct ring_buffer *buffer,
1112 unsigned long flags, int pc)
1114 struct ring_buffer_event *event;
1116 event = ring_buffer_lock_reserve(buffer, len);
1117 if (event != NULL) {
1118 struct trace_entry *ent = ring_buffer_event_data(event);
1120 tracing_generic_entry_update(ent, flags, pc);
1128 __trace_buffer_unlock_commit(struct ring_buffer *buffer,
1129 struct ring_buffer_event *event,
1130 unsigned long flags, int pc,
1133 ring_buffer_unlock_commit(buffer, event);
1135 ftrace_trace_stack(buffer, flags, 6, pc);
1136 ftrace_trace_userstack(buffer, flags, pc);
1142 void trace_buffer_unlock_commit(struct ring_buffer *buffer,
1143 struct ring_buffer_event *event,
1144 unsigned long flags, int pc)
1146 __trace_buffer_unlock_commit(buffer, event, flags, pc, 1);
1149 struct ring_buffer_event *
1150 trace_current_buffer_lock_reserve(struct ring_buffer **current_rb,
1151 int type, unsigned long len,
1152 unsigned long flags, int pc)
1154 *current_rb = global_trace.buffer;
1155 return trace_buffer_lock_reserve(*current_rb,
1156 type, len, flags, pc);
1158 EXPORT_SYMBOL_GPL(trace_current_buffer_lock_reserve);
1160 void trace_current_buffer_unlock_commit(struct ring_buffer *buffer,
1161 struct ring_buffer_event *event,
1162 unsigned long flags, int pc)
1164 __trace_buffer_unlock_commit(buffer, event, flags, pc, 1);
1166 EXPORT_SYMBOL_GPL(trace_current_buffer_unlock_commit);
1168 void trace_nowake_buffer_unlock_commit(struct ring_buffer *buffer,
1169 struct ring_buffer_event *event,
1170 unsigned long flags, int pc)
1172 __trace_buffer_unlock_commit(buffer, event, flags, pc, 0);
1174 EXPORT_SYMBOL_GPL(trace_nowake_buffer_unlock_commit);
1176 void trace_current_buffer_discard_commit(struct ring_buffer *buffer,
1177 struct ring_buffer_event *event)
1179 ring_buffer_discard_commit(buffer, event);
1181 EXPORT_SYMBOL_GPL(trace_current_buffer_discard_commit);
1184 trace_function(struct trace_array *tr,
1185 unsigned long ip, unsigned long parent_ip, unsigned long flags,
1188 struct ftrace_event_call *call = &event_function;
1189 struct ring_buffer *buffer = tr->buffer;
1190 struct ring_buffer_event *event;
1191 struct ftrace_entry *entry;
1193 /* If we are reading the ring buffer, don't trace */
1194 if (unlikely(__this_cpu_read(per_cpu_var(ftrace_cpu_disabled))))
1197 event = trace_buffer_lock_reserve(buffer, TRACE_FN, sizeof(*entry),
1201 entry = ring_buffer_event_data(event);
1203 entry->parent_ip = parent_ip;
1205 if (!filter_check_discard(call, entry, buffer, event))
1206 ring_buffer_unlock_commit(buffer, event);
1210 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
1211 unsigned long ip, unsigned long parent_ip, unsigned long flags,
1214 if (likely(!atomic_read(&data->disabled)))
1215 trace_function(tr, ip, parent_ip, flags, pc);
1218 #ifdef CONFIG_STACKTRACE
1219 static void __ftrace_trace_stack(struct ring_buffer *buffer,
1220 unsigned long flags,
1223 struct ftrace_event_call *call = &event_kernel_stack;
1224 struct ring_buffer_event *event;
1225 struct stack_entry *entry;
1226 struct stack_trace trace;
1228 event = trace_buffer_lock_reserve(buffer, TRACE_STACK,
1229 sizeof(*entry), flags, pc);
1232 entry = ring_buffer_event_data(event);
1233 memset(&entry->caller, 0, sizeof(entry->caller));
1235 trace.nr_entries = 0;
1236 trace.max_entries = FTRACE_STACK_ENTRIES;
1238 trace.entries = entry->caller;
1240 save_stack_trace(&trace);
1241 if (!filter_check_discard(call, entry, buffer, event))
1242 ring_buffer_unlock_commit(buffer, event);
1245 void ftrace_trace_stack(struct ring_buffer *buffer, unsigned long flags,
1248 if (!(trace_flags & TRACE_ITER_STACKTRACE))
1251 __ftrace_trace_stack(buffer, flags, skip, pc);
1254 void __trace_stack(struct trace_array *tr, unsigned long flags, int skip,
1257 __ftrace_trace_stack(tr->buffer, flags, skip, pc);
1261 * trace_dump_stack - record a stack back trace in the trace buffer
1263 void trace_dump_stack(void)
1265 unsigned long flags;
1267 if (tracing_disabled || tracing_selftest_running)
1270 local_save_flags(flags);
1272 /* skipping 3 traces, seems to get us at the caller of this function */
1273 __ftrace_trace_stack(global_trace.buffer, flags, 3, preempt_count());
1277 ftrace_trace_userstack(struct ring_buffer *buffer, unsigned long flags, int pc)
1279 struct ftrace_event_call *call = &event_user_stack;
1280 struct ring_buffer_event *event;
1281 struct userstack_entry *entry;
1282 struct stack_trace trace;
1284 if (!(trace_flags & TRACE_ITER_USERSTACKTRACE))
1287 event = trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
1288 sizeof(*entry), flags, pc);
1291 entry = ring_buffer_event_data(event);
1293 entry->tgid = current->tgid;
1294 memset(&entry->caller, 0, sizeof(entry->caller));
1296 trace.nr_entries = 0;
1297 trace.max_entries = FTRACE_STACK_ENTRIES;
1299 trace.entries = entry->caller;
1301 save_stack_trace_user(&trace);
1302 if (!filter_check_discard(call, entry, buffer, event))
1303 ring_buffer_unlock_commit(buffer, event);
1307 static void __trace_userstack(struct trace_array *tr, unsigned long flags)
1309 ftrace_trace_userstack(tr, flags, preempt_count());
1313 #endif /* CONFIG_STACKTRACE */
1316 ftrace_trace_special(void *__tr,
1317 unsigned long arg1, unsigned long arg2, unsigned long arg3,
1320 struct ftrace_event_call *call = &event_special;
1321 struct ring_buffer_event *event;
1322 struct trace_array *tr = __tr;
1323 struct ring_buffer *buffer = tr->buffer;
1324 struct special_entry *entry;
1326 event = trace_buffer_lock_reserve(buffer, TRACE_SPECIAL,
1327 sizeof(*entry), 0, pc);
1330 entry = ring_buffer_event_data(event);
1335 if (!filter_check_discard(call, entry, buffer, event))
1336 trace_buffer_unlock_commit(buffer, event, 0, pc);
1340 __trace_special(void *__tr, void *__data,
1341 unsigned long arg1, unsigned long arg2, unsigned long arg3)
1343 ftrace_trace_special(__tr, arg1, arg2, arg3, preempt_count());
1347 ftrace_special(unsigned long arg1, unsigned long arg2, unsigned long arg3)
1349 struct trace_array *tr = &global_trace;
1350 struct trace_array_cpu *data;
1351 unsigned long flags;
1355 if (tracing_disabled)
1358 pc = preempt_count();
1359 local_irq_save(flags);
1360 cpu = raw_smp_processor_id();
1361 data = tr->data[cpu];
1363 if (likely(atomic_inc_return(&data->disabled) == 1))
1364 ftrace_trace_special(tr, arg1, arg2, arg3, pc);
1366 atomic_dec(&data->disabled);
1367 local_irq_restore(flags);
1371 * trace_vbprintk - write binary msg to tracing buffer
1374 int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
1376 static arch_spinlock_t trace_buf_lock =
1377 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
1378 static u32 trace_buf[TRACE_BUF_SIZE];
1380 struct ftrace_event_call *call = &event_bprint;
1381 struct ring_buffer_event *event;
1382 struct ring_buffer *buffer;
1383 struct trace_array *tr = &global_trace;
1384 struct trace_array_cpu *data;
1385 struct bprint_entry *entry;
1386 unsigned long flags;
1389 int cpu, len = 0, size, pc;
1391 if (unlikely(tracing_selftest_running || tracing_disabled))
1394 /* Don't pollute graph traces with trace_vprintk internals */
1395 pause_graph_tracing();
1397 pc = preempt_count();
1398 resched = ftrace_preempt_disable();
1399 cpu = raw_smp_processor_id();
1400 data = tr->data[cpu];
1402 disable = atomic_inc_return(&data->disabled);
1403 if (unlikely(disable != 1))
1406 /* Lockdep uses trace_printk for lock tracing */
1407 local_irq_save(flags);
1408 arch_spin_lock(&trace_buf_lock);
1409 len = vbin_printf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1411 if (len > TRACE_BUF_SIZE || len < 0)
1414 size = sizeof(*entry) + sizeof(u32) * len;
1415 buffer = tr->buffer;
1416 event = trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
1420 entry = ring_buffer_event_data(event);
1424 memcpy(entry->buf, trace_buf, sizeof(u32) * len);
1425 if (!filter_check_discard(call, entry, buffer, event)) {
1426 ring_buffer_unlock_commit(buffer, event);
1427 ftrace_trace_stack(buffer, flags, 6, pc);
1431 arch_spin_unlock(&trace_buf_lock);
1432 local_irq_restore(flags);
1435 atomic_dec_return(&data->disabled);
1436 ftrace_preempt_enable(resched);
1437 unpause_graph_tracing();
1441 EXPORT_SYMBOL_GPL(trace_vbprintk);
1443 int trace_array_printk(struct trace_array *tr,
1444 unsigned long ip, const char *fmt, ...)
1449 if (!(trace_flags & TRACE_ITER_PRINTK))
1453 ret = trace_array_vprintk(tr, ip, fmt, ap);
1458 int trace_array_vprintk(struct trace_array *tr,
1459 unsigned long ip, const char *fmt, va_list args)
1461 static arch_spinlock_t trace_buf_lock = __ARCH_SPIN_LOCK_UNLOCKED;
1462 static char trace_buf[TRACE_BUF_SIZE];
1464 struct ftrace_event_call *call = &event_print;
1465 struct ring_buffer_event *event;
1466 struct ring_buffer *buffer;
1467 struct trace_array_cpu *data;
1468 int cpu, len = 0, size, pc;
1469 struct print_entry *entry;
1470 unsigned long irq_flags;
1473 if (tracing_disabled || tracing_selftest_running)
1476 pc = preempt_count();
1477 preempt_disable_notrace();
1478 cpu = raw_smp_processor_id();
1479 data = tr->data[cpu];
1481 disable = atomic_inc_return(&data->disabled);
1482 if (unlikely(disable != 1))
1485 pause_graph_tracing();
1486 raw_local_irq_save(irq_flags);
1487 arch_spin_lock(&trace_buf_lock);
1488 len = vsnprintf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1490 size = sizeof(*entry) + len + 1;
1491 buffer = tr->buffer;
1492 event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
1496 entry = ring_buffer_event_data(event);
1499 memcpy(&entry->buf, trace_buf, len);
1500 entry->buf[len] = '\0';
1501 if (!filter_check_discard(call, entry, buffer, event)) {
1502 ring_buffer_unlock_commit(buffer, event);
1503 ftrace_trace_stack(buffer, irq_flags, 6, pc);
1507 arch_spin_unlock(&trace_buf_lock);
1508 raw_local_irq_restore(irq_flags);
1509 unpause_graph_tracing();
1511 atomic_dec_return(&data->disabled);
1512 preempt_enable_notrace();
1517 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
1519 return trace_array_vprintk(&global_trace, ip, fmt, args);
1521 EXPORT_SYMBOL_GPL(trace_vprintk);
1523 enum trace_file_type {
1524 TRACE_FILE_LAT_FMT = 1,
1525 TRACE_FILE_ANNOTATE = 2,
1528 static void trace_iterator_increment(struct trace_iterator *iter)
1530 /* Don't allow ftrace to trace into the ring buffers */
1531 ftrace_disable_cpu();
1534 if (iter->buffer_iter[iter->cpu])
1535 ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
1537 ftrace_enable_cpu();
1540 static struct trace_entry *
1541 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts)
1543 struct ring_buffer_event *event;
1544 struct ring_buffer_iter *buf_iter = iter->buffer_iter[cpu];
1546 /* Don't allow ftrace to trace into the ring buffers */
1547 ftrace_disable_cpu();
1550 event = ring_buffer_iter_peek(buf_iter, ts);
1552 event = ring_buffer_peek(iter->tr->buffer, cpu, ts);
1554 ftrace_enable_cpu();
1556 return event ? ring_buffer_event_data(event) : NULL;
1559 static struct trace_entry *
1560 __find_next_entry(struct trace_iterator *iter, int *ent_cpu, u64 *ent_ts)
1562 struct ring_buffer *buffer = iter->tr->buffer;
1563 struct trace_entry *ent, *next = NULL;
1564 int cpu_file = iter->cpu_file;
1565 u64 next_ts = 0, ts;
1570 * If we are in a per_cpu trace file, don't bother by iterating over
1571 * all cpu and peek directly.
1573 if (cpu_file > TRACE_PIPE_ALL_CPU) {
1574 if (ring_buffer_empty_cpu(buffer, cpu_file))
1576 ent = peek_next_entry(iter, cpu_file, ent_ts);
1578 *ent_cpu = cpu_file;
1583 for_each_tracing_cpu(cpu) {
1585 if (ring_buffer_empty_cpu(buffer, cpu))
1588 ent = peek_next_entry(iter, cpu, &ts);
1591 * Pick the entry with the smallest timestamp:
1593 if (ent && (!next || ts < next_ts)) {
1601 *ent_cpu = next_cpu;
1609 /* Find the next real entry, without updating the iterator itself */
1610 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
1611 int *ent_cpu, u64 *ent_ts)
1613 return __find_next_entry(iter, ent_cpu, ent_ts);
1616 /* Find the next real entry, and increment the iterator to the next entry */
1617 static void *find_next_entry_inc(struct trace_iterator *iter)
1619 iter->ent = __find_next_entry(iter, &iter->cpu, &iter->ts);
1622 trace_iterator_increment(iter);
1624 return iter->ent ? iter : NULL;
1627 static void trace_consume(struct trace_iterator *iter)
1629 /* Don't allow ftrace to trace into the ring buffers */
1630 ftrace_disable_cpu();
1631 ring_buffer_consume(iter->tr->buffer, iter->cpu, &iter->ts);
1632 ftrace_enable_cpu();
1635 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1637 struct trace_iterator *iter = m->private;
1641 WARN_ON_ONCE(iter->leftover);
1645 /* can't go backwards */
1650 ent = find_next_entry_inc(iter);
1654 while (ent && iter->idx < i)
1655 ent = find_next_entry_inc(iter);
1662 static void tracing_iter_reset(struct trace_iterator *iter, int cpu)
1664 struct trace_array *tr = iter->tr;
1665 struct ring_buffer_event *event;
1666 struct ring_buffer_iter *buf_iter;
1667 unsigned long entries = 0;
1670 tr->data[cpu]->skipped_entries = 0;
1672 if (!iter->buffer_iter[cpu])
1675 buf_iter = iter->buffer_iter[cpu];
1676 ring_buffer_iter_reset(buf_iter);
1679 * We could have the case with the max latency tracers
1680 * that a reset never took place on a cpu. This is evident
1681 * by the timestamp being before the start of the buffer.
1683 while ((event = ring_buffer_iter_peek(buf_iter, &ts))) {
1684 if (ts >= iter->tr->time_start)
1687 ring_buffer_read(buf_iter, NULL);
1690 tr->data[cpu]->skipped_entries = entries;
1694 * The current tracer is copied to avoid a global locking
1697 static void *s_start(struct seq_file *m, loff_t *pos)
1699 struct trace_iterator *iter = m->private;
1700 static struct tracer *old_tracer;
1701 int cpu_file = iter->cpu_file;
1706 /* copy the tracer to avoid using a global lock all around */
1707 mutex_lock(&trace_types_lock);
1708 if (unlikely(old_tracer != current_trace && current_trace)) {
1709 old_tracer = current_trace;
1710 *iter->trace = *current_trace;
1712 mutex_unlock(&trace_types_lock);
1714 atomic_inc(&trace_record_cmdline_disabled);
1716 if (*pos != iter->pos) {
1721 ftrace_disable_cpu();
1723 if (cpu_file == TRACE_PIPE_ALL_CPU) {
1724 for_each_tracing_cpu(cpu)
1725 tracing_iter_reset(iter, cpu);
1727 tracing_iter_reset(iter, cpu_file);
1729 ftrace_enable_cpu();
1732 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
1737 * If we overflowed the seq_file before, then we want
1738 * to just reuse the trace_seq buffer again.
1744 p = s_next(m, p, &l);
1748 trace_event_read_lock();
1749 trace_access_lock(cpu_file);
1753 static void s_stop(struct seq_file *m, void *p)
1755 struct trace_iterator *iter = m->private;
1757 atomic_dec(&trace_record_cmdline_disabled);
1758 trace_access_unlock(iter->cpu_file);
1759 trace_event_read_unlock();
1762 static void print_lat_help_header(struct seq_file *m)
1764 seq_puts(m, "# _------=> CPU# \n");
1765 seq_puts(m, "# / _-----=> irqs-off \n");
1766 seq_puts(m, "# | / _----=> need-resched \n");
1767 seq_puts(m, "# || / _---=> hardirq/softirq \n");
1768 seq_puts(m, "# ||| / _--=> preempt-depth \n");
1769 seq_puts(m, "# |||| /_--=> lock-depth \n");
1770 seq_puts(m, "# |||||/ delay \n");
1771 seq_puts(m, "# cmd pid |||||| time | caller \n");
1772 seq_puts(m, "# \\ / |||||| \\ | / \n");
1775 static void print_func_help_header(struct seq_file *m)
1777 seq_puts(m, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
1778 seq_puts(m, "# | | | | |\n");
1783 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1785 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1786 struct trace_array *tr = iter->tr;
1787 struct trace_array_cpu *data = tr->data[tr->cpu];
1788 struct tracer *type = current_trace;
1789 unsigned long entries = 0;
1790 unsigned long total = 0;
1791 unsigned long count;
1792 const char *name = "preemption";
1799 for_each_tracing_cpu(cpu) {
1800 count = ring_buffer_entries_cpu(tr->buffer, cpu);
1802 * If this buffer has skipped entries, then we hold all
1803 * entries for the trace and we need to ignore the
1804 * ones before the time stamp.
1806 if (tr->data[cpu]->skipped_entries) {
1807 count -= tr->data[cpu]->skipped_entries;
1808 /* total is the same as the entries */
1812 ring_buffer_overrun_cpu(tr->buffer, cpu);
1816 seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
1818 seq_puts(m, "# -----------------------------------"
1819 "---------------------------------\n");
1820 seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
1821 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1822 nsecs_to_usecs(data->saved_latency),
1826 #if defined(CONFIG_PREEMPT_NONE)
1828 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1830 #elif defined(CONFIG_PREEMPT)
1835 /* These are reserved for later use */
1838 seq_printf(m, " #P:%d)\n", num_online_cpus());
1842 seq_puts(m, "# -----------------\n");
1843 seq_printf(m, "# | task: %.16s-%d "
1844 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1845 data->comm, data->pid, data->uid, data->nice,
1846 data->policy, data->rt_priority);
1847 seq_puts(m, "# -----------------\n");
1849 if (data->critical_start) {
1850 seq_puts(m, "# => started at: ");
1851 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1852 trace_print_seq(m, &iter->seq);
1853 seq_puts(m, "\n# => ended at: ");
1854 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1855 trace_print_seq(m, &iter->seq);
1856 seq_puts(m, "\n#\n");
1862 static void test_cpu_buff_start(struct trace_iterator *iter)
1864 struct trace_seq *s = &iter->seq;
1866 if (!(trace_flags & TRACE_ITER_ANNOTATE))
1869 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
1872 if (cpumask_test_cpu(iter->cpu, iter->started))
1875 if (iter->tr->data[iter->cpu]->skipped_entries)
1878 cpumask_set_cpu(iter->cpu, iter->started);
1880 /* Don't print started cpu buffer for the first entry of the trace */
1882 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
1886 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
1888 struct trace_seq *s = &iter->seq;
1889 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1890 struct trace_entry *entry;
1891 struct trace_event *event;
1895 test_cpu_buff_start(iter);
1897 event = ftrace_find_event(entry->type);
1899 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1900 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1901 if (!trace_print_lat_context(iter))
1904 if (!trace_print_context(iter))
1910 return event->trace(iter, sym_flags);
1912 if (!trace_seq_printf(s, "Unknown type %d\n", entry->type))
1915 return TRACE_TYPE_HANDLED;
1917 return TRACE_TYPE_PARTIAL_LINE;
1920 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
1922 struct trace_seq *s = &iter->seq;
1923 struct trace_entry *entry;
1924 struct trace_event *event;
1928 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1929 if (!trace_seq_printf(s, "%d %d %llu ",
1930 entry->pid, iter->cpu, iter->ts))
1934 event = ftrace_find_event(entry->type);
1936 return event->raw(iter, 0);
1938 if (!trace_seq_printf(s, "%d ?\n", entry->type))
1941 return TRACE_TYPE_HANDLED;
1943 return TRACE_TYPE_PARTIAL_LINE;
1946 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
1948 struct trace_seq *s = &iter->seq;
1949 unsigned char newline = '\n';
1950 struct trace_entry *entry;
1951 struct trace_event *event;
1955 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1956 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
1957 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
1958 SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
1961 event = ftrace_find_event(entry->type);
1963 enum print_line_t ret = event->hex(iter, 0);
1964 if (ret != TRACE_TYPE_HANDLED)
1968 SEQ_PUT_FIELD_RET(s, newline);
1970 return TRACE_TYPE_HANDLED;
1973 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
1975 struct trace_seq *s = &iter->seq;
1976 struct trace_entry *entry;
1977 struct trace_event *event;
1981 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1982 SEQ_PUT_FIELD_RET(s, entry->pid);
1983 SEQ_PUT_FIELD_RET(s, iter->cpu);
1984 SEQ_PUT_FIELD_RET(s, iter->ts);
1987 event = ftrace_find_event(entry->type);
1988 return event ? event->binary(iter, 0) : TRACE_TYPE_HANDLED;
1991 static int trace_empty(struct trace_iterator *iter)
1995 /* If we are looking at one CPU buffer, only check that one */
1996 if (iter->cpu_file != TRACE_PIPE_ALL_CPU) {
1997 cpu = iter->cpu_file;
1998 if (iter->buffer_iter[cpu]) {
1999 if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
2002 if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
2008 for_each_tracing_cpu(cpu) {
2009 if (iter->buffer_iter[cpu]) {
2010 if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
2013 if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
2021 /* Called with trace_event_read_lock() held. */
2022 static enum print_line_t print_trace_line(struct trace_iterator *iter)
2024 enum print_line_t ret;
2026 if (iter->trace && iter->trace->print_line) {
2027 ret = iter->trace->print_line(iter);
2028 if (ret != TRACE_TYPE_UNHANDLED)
2032 if (iter->ent->type == TRACE_BPRINT &&
2033 trace_flags & TRACE_ITER_PRINTK &&
2034 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2035 return trace_print_bprintk_msg_only(iter);
2037 if (iter->ent->type == TRACE_PRINT &&
2038 trace_flags & TRACE_ITER_PRINTK &&
2039 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2040 return trace_print_printk_msg_only(iter);
2042 if (trace_flags & TRACE_ITER_BIN)
2043 return print_bin_fmt(iter);
2045 if (trace_flags & TRACE_ITER_HEX)
2046 return print_hex_fmt(iter);
2048 if (trace_flags & TRACE_ITER_RAW)
2049 return print_raw_fmt(iter);
2051 return print_trace_fmt(iter);
2054 static int s_show(struct seq_file *m, void *v)
2056 struct trace_iterator *iter = v;
2059 if (iter->ent == NULL) {
2061 seq_printf(m, "# tracer: %s\n", iter->trace->name);
2064 if (iter->trace && iter->trace->print_header)
2065 iter->trace->print_header(m);
2066 else if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2067 /* print nothing if the buffers are empty */
2068 if (trace_empty(iter))
2070 print_trace_header(m, iter);
2071 if (!(trace_flags & TRACE_ITER_VERBOSE))
2072 print_lat_help_header(m);
2074 if (!(trace_flags & TRACE_ITER_VERBOSE))
2075 print_func_help_header(m);
2077 } else if (iter->leftover) {
2079 * If we filled the seq_file buffer earlier, we
2080 * want to just show it now.
2082 ret = trace_print_seq(m, &iter->seq);
2084 /* ret should this time be zero, but you never know */
2085 iter->leftover = ret;
2088 print_trace_line(iter);
2089 ret = trace_print_seq(m, &iter->seq);
2091 * If we overflow the seq_file buffer, then it will
2092 * ask us for this data again at start up.
2094 * ret is 0 if seq_file write succeeded.
2097 iter->leftover = ret;
2103 static const struct seq_operations tracer_seq_ops = {
2110 static struct trace_iterator *
2111 __tracing_open(struct inode *inode, struct file *file)
2113 long cpu_file = (long) inode->i_private;
2114 void *fail_ret = ERR_PTR(-ENOMEM);
2115 struct trace_iterator *iter;
2119 if (tracing_disabled)
2120 return ERR_PTR(-ENODEV);
2122 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2124 return ERR_PTR(-ENOMEM);
2127 * We make a copy of the current tracer to avoid concurrent
2128 * changes on it while we are reading.
2130 mutex_lock(&trace_types_lock);
2131 iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
2136 *iter->trace = *current_trace;
2138 if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL))
2141 if (current_trace && current_trace->print_max)
2144 iter->tr = &global_trace;
2146 mutex_init(&iter->mutex);
2147 iter->cpu_file = cpu_file;
2149 /* Notify the tracer early; before we stop tracing. */
2150 if (iter->trace && iter->trace->open)
2151 iter->trace->open(iter);
2153 /* Annotate start of buffers if we had overruns */
2154 if (ring_buffer_overruns(iter->tr->buffer))
2155 iter->iter_flags |= TRACE_FILE_ANNOTATE;
2157 /* stop the trace while dumping */
2160 if (iter->cpu_file == TRACE_PIPE_ALL_CPU) {
2161 for_each_tracing_cpu(cpu) {
2163 iter->buffer_iter[cpu] =
2164 ring_buffer_read_start(iter->tr->buffer, cpu);
2165 tracing_iter_reset(iter, cpu);
2168 cpu = iter->cpu_file;
2169 iter->buffer_iter[cpu] =
2170 ring_buffer_read_start(iter->tr->buffer, cpu);
2171 tracing_iter_reset(iter, cpu);
2174 ret = seq_open(file, &tracer_seq_ops);
2176 fail_ret = ERR_PTR(ret);
2180 m = file->private_data;
2183 mutex_unlock(&trace_types_lock);
2188 for_each_tracing_cpu(cpu) {
2189 if (iter->buffer_iter[cpu])
2190 ring_buffer_read_finish(iter->buffer_iter[cpu]);
2192 free_cpumask_var(iter->started);
2195 mutex_unlock(&trace_types_lock);
2202 int tracing_open_generic(struct inode *inode, struct file *filp)
2204 if (tracing_disabled)
2207 filp->private_data = inode->i_private;
2211 static int tracing_release(struct inode *inode, struct file *file)
2213 struct seq_file *m = (struct seq_file *)file->private_data;
2214 struct trace_iterator *iter;
2217 if (!(file->f_mode & FMODE_READ))
2222 mutex_lock(&trace_types_lock);
2223 for_each_tracing_cpu(cpu) {
2224 if (iter->buffer_iter[cpu])
2225 ring_buffer_read_finish(iter->buffer_iter[cpu]);
2228 if (iter->trace && iter->trace->close)
2229 iter->trace->close(iter);
2231 /* reenable tracing if it was previously enabled */
2233 mutex_unlock(&trace_types_lock);
2235 seq_release(inode, file);
2236 mutex_destroy(&iter->mutex);
2237 free_cpumask_var(iter->started);
2243 static int tracing_open(struct inode *inode, struct file *file)
2245 struct trace_iterator *iter;
2248 /* If this file was open for write, then erase contents */
2249 if ((file->f_mode & FMODE_WRITE) &&
2250 (file->f_flags & O_TRUNC)) {
2251 long cpu = (long) inode->i_private;
2253 if (cpu == TRACE_PIPE_ALL_CPU)
2254 tracing_reset_online_cpus(&global_trace);
2256 tracing_reset(&global_trace, cpu);
2259 if (file->f_mode & FMODE_READ) {
2260 iter = __tracing_open(inode, file);
2262 ret = PTR_ERR(iter);
2263 else if (trace_flags & TRACE_ITER_LATENCY_FMT)
2264 iter->iter_flags |= TRACE_FILE_LAT_FMT;
2270 t_next(struct seq_file *m, void *v, loff_t *pos)
2272 struct tracer *t = v;
2282 static void *t_start(struct seq_file *m, loff_t *pos)
2287 mutex_lock(&trace_types_lock);
2288 for (t = trace_types; t && l < *pos; t = t_next(m, t, &l))
2294 static void t_stop(struct seq_file *m, void *p)
2296 mutex_unlock(&trace_types_lock);
2299 static int t_show(struct seq_file *m, void *v)
2301 struct tracer *t = v;
2306 seq_printf(m, "%s", t->name);
2315 static const struct seq_operations show_traces_seq_ops = {
2322 static int show_traces_open(struct inode *inode, struct file *file)
2324 if (tracing_disabled)
2327 return seq_open(file, &show_traces_seq_ops);
2331 tracing_write_stub(struct file *filp, const char __user *ubuf,
2332 size_t count, loff_t *ppos)
2337 static const struct file_operations tracing_fops = {
2338 .open = tracing_open,
2340 .write = tracing_write_stub,
2341 .llseek = seq_lseek,
2342 .release = tracing_release,
2345 static const struct file_operations show_traces_fops = {
2346 .open = show_traces_open,
2348 .release = seq_release,
2352 * Only trace on a CPU if the bitmask is set:
2354 static cpumask_var_t tracing_cpumask;
2357 * The tracer itself will not take this lock, but still we want
2358 * to provide a consistent cpumask to user-space:
2360 static DEFINE_MUTEX(tracing_cpumask_update_lock);
2363 * Temporary storage for the character representation of the
2364 * CPU bitmask (and one more byte for the newline):
2366 static char mask_str[NR_CPUS + 1];
2369 tracing_cpumask_read(struct file *filp, char __user *ubuf,
2370 size_t count, loff_t *ppos)
2374 mutex_lock(&tracing_cpumask_update_lock);
2376 len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
2377 if (count - len < 2) {
2381 len += sprintf(mask_str + len, "\n");
2382 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
2385 mutex_unlock(&tracing_cpumask_update_lock);
2391 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
2392 size_t count, loff_t *ppos)
2395 cpumask_var_t tracing_cpumask_new;
2397 if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
2400 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
2404 mutex_lock(&tracing_cpumask_update_lock);
2406 local_irq_disable();
2407 arch_spin_lock(&ftrace_max_lock);
2408 for_each_tracing_cpu(cpu) {
2410 * Increase/decrease the disabled counter if we are
2411 * about to flip a bit in the cpumask:
2413 if (cpumask_test_cpu(cpu, tracing_cpumask) &&
2414 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2415 atomic_inc(&global_trace.data[cpu]->disabled);
2417 if (!cpumask_test_cpu(cpu, tracing_cpumask) &&
2418 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2419 atomic_dec(&global_trace.data[cpu]->disabled);
2422 arch_spin_unlock(&ftrace_max_lock);
2425 cpumask_copy(tracing_cpumask, tracing_cpumask_new);
2427 mutex_unlock(&tracing_cpumask_update_lock);
2428 free_cpumask_var(tracing_cpumask_new);
2433 free_cpumask_var(tracing_cpumask_new);
2438 static const struct file_operations tracing_cpumask_fops = {
2439 .open = tracing_open_generic,
2440 .read = tracing_cpumask_read,
2441 .write = tracing_cpumask_write,
2444 static int tracing_trace_options_show(struct seq_file *m, void *v)
2446 struct tracer_opt *trace_opts;
2450 mutex_lock(&trace_types_lock);
2451 tracer_flags = current_trace->flags->val;
2452 trace_opts = current_trace->flags->opts;
2454 for (i = 0; trace_options[i]; i++) {
2455 if (trace_flags & (1 << i))
2456 seq_printf(m, "%s\n", trace_options[i]);
2458 seq_printf(m, "no%s\n", trace_options[i]);
2461 for (i = 0; trace_opts[i].name; i++) {
2462 if (tracer_flags & trace_opts[i].bit)
2463 seq_printf(m, "%s\n", trace_opts[i].name);
2465 seq_printf(m, "no%s\n", trace_opts[i].name);
2467 mutex_unlock(&trace_types_lock);
2472 static int __set_tracer_option(struct tracer *trace,
2473 struct tracer_flags *tracer_flags,
2474 struct tracer_opt *opts, int neg)
2478 ret = trace->set_flag(tracer_flags->val, opts->bit, !neg);
2483 tracer_flags->val &= ~opts->bit;
2485 tracer_flags->val |= opts->bit;
2489 /* Try to assign a tracer specific option */
2490 static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
2492 struct tracer_flags *tracer_flags = trace->flags;
2493 struct tracer_opt *opts = NULL;
2496 for (i = 0; tracer_flags->opts[i].name; i++) {
2497 opts = &tracer_flags->opts[i];
2499 if (strcmp(cmp, opts->name) == 0)
2500 return __set_tracer_option(trace, trace->flags,
2507 static void set_tracer_flags(unsigned int mask, int enabled)
2509 /* do nothing if flag is already set */
2510 if (!!(trace_flags & mask) == !!enabled)
2514 trace_flags |= mask;
2516 trace_flags &= ~mask;
2520 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
2521 size_t cnt, loff_t *ppos)
2529 if (cnt >= sizeof(buf))
2532 if (copy_from_user(&buf, ubuf, cnt))
2536 cmp = strstrip(buf);
2538 if (strncmp(cmp, "no", 2) == 0) {
2543 for (i = 0; trace_options[i]; i++) {
2544 if (strcmp(cmp, trace_options[i]) == 0) {
2545 set_tracer_flags(1 << i, !neg);
2550 /* If no option could be set, test the specific tracer options */
2551 if (!trace_options[i]) {
2552 mutex_lock(&trace_types_lock);
2553 ret = set_tracer_option(current_trace, cmp, neg);
2554 mutex_unlock(&trace_types_lock);
2564 static int tracing_trace_options_open(struct inode *inode, struct file *file)
2566 if (tracing_disabled)
2568 return single_open(file, tracing_trace_options_show, NULL);
2571 static const struct file_operations tracing_iter_fops = {
2572 .open = tracing_trace_options_open,
2574 .llseek = seq_lseek,
2575 .release = single_release,
2576 .write = tracing_trace_options_write,
2579 static const char readme_msg[] =
2580 "tracing mini-HOWTO:\n\n"
2581 "# mount -t debugfs nodev /sys/kernel/debug\n\n"
2582 "# cat /sys/kernel/debug/tracing/available_tracers\n"
2583 "wakeup preemptirqsoff preemptoff irqsoff function sched_switch nop\n\n"
2584 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2586 "# echo sched_switch > /sys/kernel/debug/tracing/current_tracer\n"
2587 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2589 "# cat /sys/kernel/debug/tracing/trace_options\n"
2590 "noprint-parent nosym-offset nosym-addr noverbose\n"
2591 "# echo print-parent > /sys/kernel/debug/tracing/trace_options\n"
2592 "# echo 1 > /sys/kernel/debug/tracing/tracing_enabled\n"
2593 "# cat /sys/kernel/debug/tracing/trace > /tmp/trace.txt\n"
2594 "# echo 0 > /sys/kernel/debug/tracing/tracing_enabled\n"
2598 tracing_readme_read(struct file *filp, char __user *ubuf,
2599 size_t cnt, loff_t *ppos)
2601 return simple_read_from_buffer(ubuf, cnt, ppos,
2602 readme_msg, strlen(readme_msg));
2605 static const struct file_operations tracing_readme_fops = {
2606 .open = tracing_open_generic,
2607 .read = tracing_readme_read,
2611 tracing_saved_cmdlines_read(struct file *file, char __user *ubuf,
2612 size_t cnt, loff_t *ppos)
2621 file_buf = kmalloc(SAVED_CMDLINES*(16+TASK_COMM_LEN), GFP_KERNEL);
2625 buf_comm = kmalloc(TASK_COMM_LEN, GFP_KERNEL);
2633 for (i = 0; i < SAVED_CMDLINES; i++) {
2636 pid = map_cmdline_to_pid[i];
2637 if (pid == -1 || pid == NO_CMDLINE_MAP)
2640 trace_find_cmdline(pid, buf_comm);
2641 r = sprintf(buf, "%d %s\n", pid, buf_comm);
2646 len = simple_read_from_buffer(ubuf, cnt, ppos,
2655 static const struct file_operations tracing_saved_cmdlines_fops = {
2656 .open = tracing_open_generic,
2657 .read = tracing_saved_cmdlines_read,
2661 tracing_ctrl_read(struct file *filp, char __user *ubuf,
2662 size_t cnt, loff_t *ppos)
2667 r = sprintf(buf, "%u\n", tracer_enabled);
2668 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2672 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2673 size_t cnt, loff_t *ppos)
2675 struct trace_array *tr = filp->private_data;
2680 if (cnt >= sizeof(buf))
2683 if (copy_from_user(&buf, ubuf, cnt))
2688 ret = strict_strtoul(buf, 10, &val);
2694 mutex_lock(&trace_types_lock);
2695 if (tracer_enabled ^ val) {
2698 if (current_trace->start)
2699 current_trace->start(tr);
2704 if (current_trace->stop)
2705 current_trace->stop(tr);
2708 mutex_unlock(&trace_types_lock);
2716 tracing_set_trace_read(struct file *filp, char __user *ubuf,
2717 size_t cnt, loff_t *ppos)
2719 char buf[MAX_TRACER_SIZE+2];
2722 mutex_lock(&trace_types_lock);
2724 r = sprintf(buf, "%s\n", current_trace->name);
2726 r = sprintf(buf, "\n");
2727 mutex_unlock(&trace_types_lock);
2729 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2732 int tracer_init(struct tracer *t, struct trace_array *tr)
2734 tracing_reset_online_cpus(tr);
2738 static int tracing_resize_ring_buffer(unsigned long size)
2743 * If kernel or user changes the size of the ring buffer
2744 * we use the size that was given, and we can forget about
2745 * expanding it later.
2747 ring_buffer_expanded = 1;
2749 ret = ring_buffer_resize(global_trace.buffer, size);
2753 ret = ring_buffer_resize(max_tr.buffer, size);
2757 r = ring_buffer_resize(global_trace.buffer,
2758 global_trace.entries);
2761 * AARGH! We are left with different
2762 * size max buffer!!!!
2763 * The max buffer is our "snapshot" buffer.
2764 * When a tracer needs a snapshot (one of the
2765 * latency tracers), it swaps the max buffer
2766 * with the saved snap shot. We succeeded to
2767 * update the size of the main buffer, but failed to
2768 * update the size of the max buffer. But when we tried
2769 * to reset the main buffer to the original size, we
2770 * failed there too. This is very unlikely to
2771 * happen, but if it does, warn and kill all
2775 tracing_disabled = 1;
2780 global_trace.entries = size;
2786 * tracing_update_buffers - used by tracing facility to expand ring buffers
2788 * To save on memory when the tracing is never used on a system with it
2789 * configured in. The ring buffers are set to a minimum size. But once
2790 * a user starts to use the tracing facility, then they need to grow
2791 * to their default size.
2793 * This function is to be called when a tracer is about to be used.
2795 int tracing_update_buffers(void)
2799 mutex_lock(&trace_types_lock);
2800 if (!ring_buffer_expanded)
2801 ret = tracing_resize_ring_buffer(trace_buf_size);
2802 mutex_unlock(&trace_types_lock);
2807 struct trace_option_dentry;
2809 static struct trace_option_dentry *
2810 create_trace_option_files(struct tracer *tracer);
2813 destroy_trace_option_files(struct trace_option_dentry *topts);
2815 static int tracing_set_tracer(const char *buf)
2817 static struct trace_option_dentry *topts;
2818 struct trace_array *tr = &global_trace;
2822 mutex_lock(&trace_types_lock);
2824 if (!ring_buffer_expanded) {
2825 ret = tracing_resize_ring_buffer(trace_buf_size);
2831 for (t = trace_types; t; t = t->next) {
2832 if (strcmp(t->name, buf) == 0)
2839 if (t == current_trace)
2842 trace_branch_disable();
2843 if (current_trace && current_trace->reset)
2844 current_trace->reset(tr);
2846 destroy_trace_option_files(topts);
2850 topts = create_trace_option_files(current_trace);
2853 ret = tracer_init(t, tr);
2858 trace_branch_enable(tr);
2860 mutex_unlock(&trace_types_lock);
2866 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
2867 size_t cnt, loff_t *ppos)
2869 char buf[MAX_TRACER_SIZE+1];
2876 if (cnt > MAX_TRACER_SIZE)
2877 cnt = MAX_TRACER_SIZE;
2879 if (copy_from_user(&buf, ubuf, cnt))
2884 /* strip ending whitespace. */
2885 for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
2888 err = tracing_set_tracer(buf);
2898 tracing_max_lat_read(struct file *filp, char __user *ubuf,
2899 size_t cnt, loff_t *ppos)
2901 unsigned long *ptr = filp->private_data;
2905 r = snprintf(buf, sizeof(buf), "%ld\n",
2906 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
2907 if (r > sizeof(buf))
2909 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2913 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
2914 size_t cnt, loff_t *ppos)
2916 unsigned long *ptr = filp->private_data;
2921 if (cnt >= sizeof(buf))
2924 if (copy_from_user(&buf, ubuf, cnt))
2929 ret = strict_strtoul(buf, 10, &val);
2938 static int tracing_open_pipe(struct inode *inode, struct file *filp)
2940 long cpu_file = (long) inode->i_private;
2941 struct trace_iterator *iter;
2944 if (tracing_disabled)
2947 mutex_lock(&trace_types_lock);
2949 /* create a buffer to store the information to pass to userspace */
2950 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2957 * We make a copy of the current tracer to avoid concurrent
2958 * changes on it while we are reading.
2960 iter->trace = kmalloc(sizeof(*iter->trace), GFP_KERNEL);
2966 *iter->trace = *current_trace;
2968 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
2973 /* trace pipe does not show start of buffer */
2974 cpumask_setall(iter->started);
2976 if (trace_flags & TRACE_ITER_LATENCY_FMT)
2977 iter->iter_flags |= TRACE_FILE_LAT_FMT;
2979 iter->cpu_file = cpu_file;
2980 iter->tr = &global_trace;
2981 mutex_init(&iter->mutex);
2982 filp->private_data = iter;
2984 if (iter->trace->pipe_open)
2985 iter->trace->pipe_open(iter);
2988 mutex_unlock(&trace_types_lock);
2994 mutex_unlock(&trace_types_lock);
2998 static int tracing_release_pipe(struct inode *inode, struct file *file)
3000 struct trace_iterator *iter = file->private_data;
3002 mutex_lock(&trace_types_lock);
3004 if (iter->trace->pipe_close)
3005 iter->trace->pipe_close(iter);
3007 mutex_unlock(&trace_types_lock);
3009 free_cpumask_var(iter->started);
3010 mutex_destroy(&iter->mutex);
3018 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
3020 struct trace_iterator *iter = filp->private_data;
3022 if (trace_flags & TRACE_ITER_BLOCK) {
3024 * Always select as readable when in blocking mode
3026 return POLLIN | POLLRDNORM;
3028 if (!trace_empty(iter))
3029 return POLLIN | POLLRDNORM;
3030 poll_wait(filp, &trace_wait, poll_table);
3031 if (!trace_empty(iter))
3032 return POLLIN | POLLRDNORM;
3039 void default_wait_pipe(struct trace_iterator *iter)
3043 prepare_to_wait(&trace_wait, &wait, TASK_INTERRUPTIBLE);
3045 if (trace_empty(iter))
3048 finish_wait(&trace_wait, &wait);
3052 * This is a make-shift waitqueue.
3053 * A tracer might use this callback on some rare cases:
3055 * 1) the current tracer might hold the runqueue lock when it wakes up
3056 * a reader, hence a deadlock (sched, function, and function graph tracers)
3057 * 2) the function tracers, trace all functions, we don't want
3058 * the overhead of calling wake_up and friends
3059 * (and tracing them too)
3061 * Anyway, this is really very primitive wakeup.
3063 void poll_wait_pipe(struct trace_iterator *iter)
3065 set_current_state(TASK_INTERRUPTIBLE);
3066 /* sleep for 100 msecs, and try again. */
3067 schedule_timeout(HZ / 10);
3070 /* Must be called with trace_types_lock mutex held. */
3071 static int tracing_wait_pipe(struct file *filp)
3073 struct trace_iterator *iter = filp->private_data;
3075 while (trace_empty(iter)) {
3077 if ((filp->f_flags & O_NONBLOCK)) {
3081 mutex_unlock(&iter->mutex);
3083 iter->trace->wait_pipe(iter);
3085 mutex_lock(&iter->mutex);
3087 if (signal_pending(current))
3091 * We block until we read something and tracing is disabled.
3092 * We still block if tracing is disabled, but we have never
3093 * read anything. This allows a user to cat this file, and
3094 * then enable tracing. But after we have read something,
3095 * we give an EOF when tracing is again disabled.
3097 * iter->pos will be 0 if we haven't read anything.
3099 if (!tracer_enabled && iter->pos)
3110 tracing_read_pipe(struct file *filp, char __user *ubuf,
3111 size_t cnt, loff_t *ppos)
3113 struct trace_iterator *iter = filp->private_data;
3114 static struct tracer *old_tracer;
3117 /* return any leftover data */
3118 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3122 trace_seq_init(&iter->seq);
3124 /* copy the tracer to avoid using a global lock all around */
3125 mutex_lock(&trace_types_lock);
3126 if (unlikely(old_tracer != current_trace && current_trace)) {
3127 old_tracer = current_trace;
3128 *iter->trace = *current_trace;
3130 mutex_unlock(&trace_types_lock);
3133 * Avoid more than one consumer on a single file descriptor
3134 * This is just a matter of traces coherency, the ring buffer itself
3137 mutex_lock(&iter->mutex);
3138 if (iter->trace->read) {
3139 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
3145 sret = tracing_wait_pipe(filp);
3149 /* stop when tracing is finished */
3150 if (trace_empty(iter)) {
3155 if (cnt >= PAGE_SIZE)
3156 cnt = PAGE_SIZE - 1;
3158 /* reset all but tr, trace, and overruns */
3159 memset(&iter->seq, 0,
3160 sizeof(struct trace_iterator) -
3161 offsetof(struct trace_iterator, seq));
3164 trace_event_read_lock();
3165 trace_access_lock(iter->cpu_file);
3166 while (find_next_entry_inc(iter) != NULL) {
3167 enum print_line_t ret;
3168 int len = iter->seq.len;
3170 ret = print_trace_line(iter);
3171 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3172 /* don't print partial lines */
3173 iter->seq.len = len;
3176 if (ret != TRACE_TYPE_NO_CONSUME)
3177 trace_consume(iter);
3179 if (iter->seq.len >= cnt)
3182 trace_access_unlock(iter->cpu_file);
3183 trace_event_read_unlock();
3185 /* Now copy what we have to the user */
3186 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3187 if (iter->seq.readpos >= iter->seq.len)
3188 trace_seq_init(&iter->seq);
3191 * If there was nothing to send to user, inspite of consuming trace
3192 * entries, go back to wait for more entries.
3198 mutex_unlock(&iter->mutex);
3203 static void tracing_pipe_buf_release(struct pipe_inode_info *pipe,
3204 struct pipe_buffer *buf)
3206 __free_page(buf->page);
3209 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
3212 __free_page(spd->pages[idx]);
3215 static const struct pipe_buf_operations tracing_pipe_buf_ops = {
3217 .map = generic_pipe_buf_map,
3218 .unmap = generic_pipe_buf_unmap,
3219 .confirm = generic_pipe_buf_confirm,
3220 .release = tracing_pipe_buf_release,
3221 .steal = generic_pipe_buf_steal,
3222 .get = generic_pipe_buf_get,
3226 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
3231 /* Seq buffer is page-sized, exactly what we need. */
3233 count = iter->seq.len;
3234 ret = print_trace_line(iter);
3235 count = iter->seq.len - count;
3238 iter->seq.len -= count;
3241 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3242 iter->seq.len -= count;
3246 if (ret != TRACE_TYPE_NO_CONSUME)
3247 trace_consume(iter);
3249 if (!find_next_entry_inc(iter)) {
3259 static ssize_t tracing_splice_read_pipe(struct file *filp,
3261 struct pipe_inode_info *pipe,
3265 struct page *pages[PIPE_BUFFERS];
3266 struct partial_page partial[PIPE_BUFFERS];
3267 struct trace_iterator *iter = filp->private_data;
3268 struct splice_pipe_desc spd = {
3271 .nr_pages = 0, /* This gets updated below. */
3273 .ops = &tracing_pipe_buf_ops,
3274 .spd_release = tracing_spd_release_pipe,
3276 static struct tracer *old_tracer;
3281 /* copy the tracer to avoid using a global lock all around */
3282 mutex_lock(&trace_types_lock);
3283 if (unlikely(old_tracer != current_trace && current_trace)) {
3284 old_tracer = current_trace;
3285 *iter->trace = *current_trace;
3287 mutex_unlock(&trace_types_lock);
3289 mutex_lock(&iter->mutex);
3291 if (iter->trace->splice_read) {
3292 ret = iter->trace->splice_read(iter, filp,
3293 ppos, pipe, len, flags);
3298 ret = tracing_wait_pipe(filp);
3302 if (!iter->ent && !find_next_entry_inc(iter)) {
3307 trace_event_read_lock();
3308 trace_access_lock(iter->cpu_file);
3310 /* Fill as many pages as possible. */
3311 for (i = 0, rem = len; i < PIPE_BUFFERS && rem; i++) {
3312 pages[i] = alloc_page(GFP_KERNEL);
3316 rem = tracing_fill_pipe_page(rem, iter);
3318 /* Copy the data into the page, so we can start over. */
3319 ret = trace_seq_to_buffer(&iter->seq,
3320 page_address(pages[i]),
3323 __free_page(pages[i]);
3326 partial[i].offset = 0;
3327 partial[i].len = iter->seq.len;
3329 trace_seq_init(&iter->seq);
3332 trace_access_unlock(iter->cpu_file);
3333 trace_event_read_unlock();
3334 mutex_unlock(&iter->mutex);
3338 return splice_to_pipe(pipe, &spd);
3341 mutex_unlock(&iter->mutex);
3347 tracing_entries_read(struct file *filp, char __user *ubuf,
3348 size_t cnt, loff_t *ppos)
3350 struct trace_array *tr = filp->private_data;
3354 mutex_lock(&trace_types_lock);
3355 if (!ring_buffer_expanded)
3356 r = sprintf(buf, "%lu (expanded: %lu)\n",
3358 trace_buf_size >> 10);
3360 r = sprintf(buf, "%lu\n", tr->entries >> 10);
3361 mutex_unlock(&trace_types_lock);
3363 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3367 tracing_entries_write(struct file *filp, const char __user *ubuf,
3368 size_t cnt, loff_t *ppos)
3374 if (cnt >= sizeof(buf))
3377 if (copy_from_user(&buf, ubuf, cnt))
3382 ret = strict_strtoul(buf, 10, &val);
3386 /* must have at least 1 entry */
3390 mutex_lock(&trace_types_lock);
3394 /* disable all cpu buffers */
3395 for_each_tracing_cpu(cpu) {
3396 if (global_trace.data[cpu])
3397 atomic_inc(&global_trace.data[cpu]->disabled);
3398 if (max_tr.data[cpu])
3399 atomic_inc(&max_tr.data[cpu]->disabled);
3402 /* value is in KB */
3405 if (val != global_trace.entries) {
3406 ret = tracing_resize_ring_buffer(val);
3415 /* If check pages failed, return ENOMEM */
3416 if (tracing_disabled)
3419 for_each_tracing_cpu(cpu) {
3420 if (global_trace.data[cpu])
3421 atomic_dec(&global_trace.data[cpu]->disabled);
3422 if (max_tr.data[cpu])
3423 atomic_dec(&max_tr.data[cpu]->disabled);
3427 max_tr.entries = global_trace.entries;
3428 mutex_unlock(&trace_types_lock);
3433 static int mark_printk(const char *fmt, ...)
3437 va_start(args, fmt);
3438 ret = trace_vprintk(0, fmt, args);
3444 tracing_mark_write(struct file *filp, const char __user *ubuf,
3445 size_t cnt, loff_t *fpos)
3449 if (tracing_disabled)
3452 if (cnt > TRACE_BUF_SIZE)
3453 cnt = TRACE_BUF_SIZE;
3455 buf = kmalloc(cnt + 2, GFP_KERNEL);
3459 if (copy_from_user(buf, ubuf, cnt)) {
3463 if (buf[cnt-1] != '\n') {
3469 cnt = mark_printk("%s", buf);
3476 static int tracing_clock_show(struct seq_file *m, void *v)
3480 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)
3482 "%s%s%s%s", i ? " " : "",
3483 i == trace_clock_id ? "[" : "", trace_clocks[i].name,
3484 i == trace_clock_id ? "]" : "");
3490 static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf,
3491 size_t cnt, loff_t *fpos)
3494 const char *clockstr;
3497 if (cnt >= sizeof(buf))
3500 if (copy_from_user(&buf, ubuf, cnt))
3505 clockstr = strstrip(buf);
3507 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) {
3508 if (strcmp(trace_clocks[i].name, clockstr) == 0)
3511 if (i == ARRAY_SIZE(trace_clocks))
3516 mutex_lock(&trace_types_lock);
3518 ring_buffer_set_clock(global_trace.buffer, trace_clocks[i].func);
3520 ring_buffer_set_clock(max_tr.buffer, trace_clocks[i].func);
3522 mutex_unlock(&trace_types_lock);
3529 static int tracing_clock_open(struct inode *inode, struct file *file)
3531 if (tracing_disabled)
3533 return single_open(file, tracing_clock_show, NULL);
3536 static const struct file_operations tracing_max_lat_fops = {
3537 .open = tracing_open_generic,
3538 .read = tracing_max_lat_read,
3539 .write = tracing_max_lat_write,
3542 static const struct file_operations tracing_ctrl_fops = {
3543 .open = tracing_open_generic,
3544 .read = tracing_ctrl_read,
3545 .write = tracing_ctrl_write,
3548 static const struct file_operations set_tracer_fops = {
3549 .open = tracing_open_generic,
3550 .read = tracing_set_trace_read,
3551 .write = tracing_set_trace_write,
3554 static const struct file_operations tracing_pipe_fops = {
3555 .open = tracing_open_pipe,
3556 .poll = tracing_poll_pipe,
3557 .read = tracing_read_pipe,
3558 .splice_read = tracing_splice_read_pipe,
3559 .release = tracing_release_pipe,
3562 static const struct file_operations tracing_entries_fops = {
3563 .open = tracing_open_generic,
3564 .read = tracing_entries_read,
3565 .write = tracing_entries_write,
3568 static const struct file_operations tracing_mark_fops = {
3569 .open = tracing_open_generic,
3570 .write = tracing_mark_write,
3573 static const struct file_operations trace_clock_fops = {
3574 .open = tracing_clock_open,
3576 .llseek = seq_lseek,
3577 .release = single_release,
3578 .write = tracing_clock_write,
3581 struct ftrace_buffer_info {
3582 struct trace_array *tr;
3588 static int tracing_buffers_open(struct inode *inode, struct file *filp)
3590 int cpu = (int)(long)inode->i_private;
3591 struct ftrace_buffer_info *info;
3593 if (tracing_disabled)
3596 info = kzalloc(sizeof(*info), GFP_KERNEL);
3600 info->tr = &global_trace;
3603 /* Force reading ring buffer for first read */
3604 info->read = (unsigned int)-1;
3606 filp->private_data = info;
3608 return nonseekable_open(inode, filp);
3612 tracing_buffers_read(struct file *filp, char __user *ubuf,
3613 size_t count, loff_t *ppos)
3615 struct ftrace_buffer_info *info = filp->private_data;
3624 info->spare = ring_buffer_alloc_read_page(info->tr->buffer);
3628 /* Do we have previous read data to read? */
3629 if (info->read < PAGE_SIZE)
3634 trace_access_lock(info->cpu);
3635 ret = ring_buffer_read_page(info->tr->buffer,
3639 trace_access_unlock(info->cpu);
3643 pos = ring_buffer_page_len(info->spare);
3645 if (pos < PAGE_SIZE)
3646 memset(info->spare + pos, 0, PAGE_SIZE - pos);
3649 size = PAGE_SIZE - info->read;
3653 ret = copy_to_user(ubuf, info->spare + info->read, size);
3664 static int tracing_buffers_release(struct inode *inode, struct file *file)
3666 struct ftrace_buffer_info *info = file->private_data;
3669 ring_buffer_free_read_page(info->tr->buffer, info->spare);
3676 struct ring_buffer *buffer;
3681 static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
3682 struct pipe_buffer *buf)
3684 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3689 ring_buffer_free_read_page(ref->buffer, ref->page);
3694 static int buffer_pipe_buf_steal(struct pipe_inode_info *pipe,
3695 struct pipe_buffer *buf)
3700 static void buffer_pipe_buf_get(struct pipe_inode_info *pipe,
3701 struct pipe_buffer *buf)
3703 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3708 /* Pipe buffer operations for a buffer. */
3709 static const struct pipe_buf_operations buffer_pipe_buf_ops = {
3711 .map = generic_pipe_buf_map,
3712 .unmap = generic_pipe_buf_unmap,
3713 .confirm = generic_pipe_buf_confirm,
3714 .release = buffer_pipe_buf_release,
3715 .steal = buffer_pipe_buf_steal,
3716 .get = buffer_pipe_buf_get,
3720 * Callback from splice_to_pipe(), if we need to release some pages
3721 * at the end of the spd in case we error'ed out in filling the pipe.
3723 static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
3725 struct buffer_ref *ref =
3726 (struct buffer_ref *)spd->partial[i].private;
3731 ring_buffer_free_read_page(ref->buffer, ref->page);
3733 spd->partial[i].private = 0;
3737 tracing_buffers_splice_read(struct file *file, loff_t *ppos,
3738 struct pipe_inode_info *pipe, size_t len,
3741 struct ftrace_buffer_info *info = file->private_data;
3742 struct partial_page partial[PIPE_BUFFERS];
3743 struct page *pages[PIPE_BUFFERS];
3744 struct splice_pipe_desc spd = {
3748 .ops = &buffer_pipe_buf_ops,
3749 .spd_release = buffer_spd_release,
3751 struct buffer_ref *ref;
3752 int entries, size, i;
3755 if (*ppos & (PAGE_SIZE - 1)) {
3756 WARN_ONCE(1, "Ftrace: previous read must page-align\n");
3760 if (len & (PAGE_SIZE - 1)) {
3761 WARN_ONCE(1, "Ftrace: splice_read should page-align\n");
3762 if (len < PAGE_SIZE)
3767 trace_access_lock(info->cpu);
3768 entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
3770 for (i = 0; i < PIPE_BUFFERS && len && entries; i++, len -= PAGE_SIZE) {
3774 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
3779 ref->buffer = info->tr->buffer;
3780 ref->page = ring_buffer_alloc_read_page(ref->buffer);
3786 r = ring_buffer_read_page(ref->buffer, &ref->page,
3789 ring_buffer_free_read_page(ref->buffer,
3796 * zero out any left over data, this is going to
3799 size = ring_buffer_page_len(ref->page);
3800 if (size < PAGE_SIZE)
3801 memset(ref->page + size, 0, PAGE_SIZE - size);
3803 page = virt_to_page(ref->page);
3805 spd.pages[i] = page;
3806 spd.partial[i].len = PAGE_SIZE;
3807 spd.partial[i].offset = 0;
3808 spd.partial[i].private = (unsigned long)ref;
3812 entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
3815 trace_access_unlock(info->cpu);
3818 /* did we read anything? */
3819 if (!spd.nr_pages) {
3820 if (flags & SPLICE_F_NONBLOCK)
3828 ret = splice_to_pipe(pipe, &spd);
3833 static const struct file_operations tracing_buffers_fops = {
3834 .open = tracing_buffers_open,
3835 .read = tracing_buffers_read,
3836 .release = tracing_buffers_release,
3837 .splice_read = tracing_buffers_splice_read,
3838 .llseek = no_llseek,
3842 tracing_stats_read(struct file *filp, char __user *ubuf,
3843 size_t count, loff_t *ppos)
3845 unsigned long cpu = (unsigned long)filp->private_data;
3846 struct trace_array *tr = &global_trace;
3847 struct trace_seq *s;
3850 s = kmalloc(sizeof(*s), GFP_KERNEL);
3856 cnt = ring_buffer_entries_cpu(tr->buffer, cpu);
3857 trace_seq_printf(s, "entries: %ld\n", cnt);
3859 cnt = ring_buffer_overrun_cpu(tr->buffer, cpu);
3860 trace_seq_printf(s, "overrun: %ld\n", cnt);
3862 cnt = ring_buffer_commit_overrun_cpu(tr->buffer, cpu);
3863 trace_seq_printf(s, "commit overrun: %ld\n", cnt);
3865 count = simple_read_from_buffer(ubuf, count, ppos, s->buffer, s->len);
3872 static const struct file_operations tracing_stats_fops = {
3873 .open = tracing_open_generic,
3874 .read = tracing_stats_read,
3877 #ifdef CONFIG_DYNAMIC_FTRACE
3879 int __weak ftrace_arch_read_dyn_info(char *buf, int size)
3885 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
3886 size_t cnt, loff_t *ppos)
3888 static char ftrace_dyn_info_buffer[1024];
3889 static DEFINE_MUTEX(dyn_info_mutex);
3890 unsigned long *p = filp->private_data;
3891 char *buf = ftrace_dyn_info_buffer;
3892 int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
3895 mutex_lock(&dyn_info_mutex);
3896 r = sprintf(buf, "%ld ", *p);
3898 r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
3901 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3903 mutex_unlock(&dyn_info_mutex);
3908 static const struct file_operations tracing_dyn_info_fops = {
3909 .open = tracing_open_generic,
3910 .read = tracing_read_dyn_info,
3914 static struct dentry *d_tracer;
3916 struct dentry *tracing_init_dentry(void)
3923 if (!debugfs_initialized())
3926 d_tracer = debugfs_create_dir("tracing", NULL);
3928 if (!d_tracer && !once) {
3930 pr_warning("Could not create debugfs directory 'tracing'\n");
3937 static struct dentry *d_percpu;
3939 struct dentry *tracing_dentry_percpu(void)
3942 struct dentry *d_tracer;
3947 d_tracer = tracing_init_dentry();
3952 d_percpu = debugfs_create_dir("per_cpu", d_tracer);
3954 if (!d_percpu && !once) {
3956 pr_warning("Could not create debugfs directory 'per_cpu'\n");
3963 static void tracing_init_debugfs_percpu(long cpu)
3965 struct dentry *d_percpu = tracing_dentry_percpu();
3966 struct dentry *d_cpu;
3967 /* strlen(cpu) + MAX(log10(cpu)) + '\0' */
3970 if (cpu > 999 || cpu < 0)
3973 sprintf(cpu_dir, "cpu%ld", cpu);
3974 d_cpu = debugfs_create_dir(cpu_dir, d_percpu);
3976 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir);
3980 /* per cpu trace_pipe */
3981 trace_create_file("trace_pipe", 0444, d_cpu,
3982 (void *) cpu, &tracing_pipe_fops);
3985 trace_create_file("trace", 0644, d_cpu,
3986 (void *) cpu, &tracing_fops);
3988 trace_create_file("trace_pipe_raw", 0444, d_cpu,
3989 (void *) cpu, &tracing_buffers_fops);
3991 trace_create_file("stats", 0444, d_cpu,
3992 (void *) cpu, &tracing_stats_fops);
3995 #ifdef CONFIG_FTRACE_SELFTEST
3996 /* Let selftest have access to static functions in this file */
3997 #include "trace_selftest.c"
4000 struct trace_option_dentry {
4001 struct tracer_opt *opt;
4002 struct tracer_flags *flags;
4003 struct dentry *entry;
4007 trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
4010 struct trace_option_dentry *topt = filp->private_data;
4013 if (topt->flags->val & topt->opt->bit)
4018 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
4022 trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
4025 struct trace_option_dentry *topt = filp->private_data;
4030 if (cnt >= sizeof(buf))
4033 if (copy_from_user(&buf, ubuf, cnt))
4038 ret = strict_strtoul(buf, 10, &val);
4042 if (val != 0 && val != 1)
4045 if (!!(topt->flags->val & topt->opt->bit) != val) {
4046 mutex_lock(&trace_types_lock);
4047 ret = __set_tracer_option(current_trace, topt->flags,
4049 mutex_unlock(&trace_types_lock);
4060 static const struct file_operations trace_options_fops = {
4061 .open = tracing_open_generic,
4062 .read = trace_options_read,
4063 .write = trace_options_write,
4067 trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
4070 long index = (long)filp->private_data;
4073 if (trace_flags & (1 << index))
4078 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
4082 trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
4085 long index = (long)filp->private_data;
4090 if (cnt >= sizeof(buf))
4093 if (copy_from_user(&buf, ubuf, cnt))
4098 ret = strict_strtoul(buf, 10, &val);
4102 if (val != 0 && val != 1)
4104 set_tracer_flags(1 << index, val);
4111 static const struct file_operations trace_options_core_fops = {
4112 .open = tracing_open_generic,
4113 .read = trace_options_core_read,
4114 .write = trace_options_core_write,
4117 struct dentry *trace_create_file(const char *name,
4119 struct dentry *parent,
4121 const struct file_operations *fops)
4125 ret = debugfs_create_file(name, mode, parent, data, fops);
4127 pr_warning("Could not create debugfs '%s' entry\n", name);
4133 static struct dentry *trace_options_init_dentry(void)
4135 struct dentry *d_tracer;
4136 static struct dentry *t_options;
4141 d_tracer = tracing_init_dentry();
4145 t_options = debugfs_create_dir("options", d_tracer);
4147 pr_warning("Could not create debugfs directory 'options'\n");
4155 create_trace_option_file(struct trace_option_dentry *topt,
4156 struct tracer_flags *flags,
4157 struct tracer_opt *opt)
4159 struct dentry *t_options;
4161 t_options = trace_options_init_dentry();
4165 topt->flags = flags;
4168 topt->entry = trace_create_file(opt->name, 0644, t_options, topt,
4169 &trace_options_fops);
4173 static struct trace_option_dentry *
4174 create_trace_option_files(struct tracer *tracer)
4176 struct trace_option_dentry *topts;
4177 struct tracer_flags *flags;
4178 struct tracer_opt *opts;
4184 flags = tracer->flags;
4186 if (!flags || !flags->opts)
4191 for (cnt = 0; opts[cnt].name; cnt++)
4194 topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
4198 for (cnt = 0; opts[cnt].name; cnt++)
4199 create_trace_option_file(&topts[cnt], flags,
4206 destroy_trace_option_files(struct trace_option_dentry *topts)
4213 for (cnt = 0; topts[cnt].opt; cnt++) {
4214 if (topts[cnt].entry)
4215 debugfs_remove(topts[cnt].entry);
4221 static struct dentry *
4222 create_trace_option_core_file(const char *option, long index)
4224 struct dentry *t_options;
4226 t_options = trace_options_init_dentry();
4230 return trace_create_file(option, 0644, t_options, (void *)index,
4231 &trace_options_core_fops);
4234 static __init void create_trace_options_dir(void)
4236 struct dentry *t_options;
4239 t_options = trace_options_init_dentry();
4243 for (i = 0; trace_options[i]; i++)
4244 create_trace_option_core_file(trace_options[i], i);
4247 static __init int tracer_init_debugfs(void)
4249 struct dentry *d_tracer;
4252 trace_access_lock_init();
4254 d_tracer = tracing_init_dentry();
4256 trace_create_file("tracing_enabled", 0644, d_tracer,
4257 &global_trace, &tracing_ctrl_fops);
4259 trace_create_file("trace_options", 0644, d_tracer,
4260 NULL, &tracing_iter_fops);
4262 trace_create_file("tracing_cpumask", 0644, d_tracer,
4263 NULL, &tracing_cpumask_fops);
4265 trace_create_file("trace", 0644, d_tracer,
4266 (void *) TRACE_PIPE_ALL_CPU, &tracing_fops);
4268 trace_create_file("available_tracers", 0444, d_tracer,
4269 &global_trace, &show_traces_fops);
4271 trace_create_file("current_tracer", 0644, d_tracer,
4272 &global_trace, &set_tracer_fops);
4274 #ifdef CONFIG_TRACER_MAX_TRACE
4275 trace_create_file("tracing_max_latency", 0644, d_tracer,
4276 &tracing_max_latency, &tracing_max_lat_fops);
4279 trace_create_file("tracing_thresh", 0644, d_tracer,
4280 &tracing_thresh, &tracing_max_lat_fops);
4282 trace_create_file("README", 0444, d_tracer,
4283 NULL, &tracing_readme_fops);
4285 trace_create_file("trace_pipe", 0444, d_tracer,
4286 (void *) TRACE_PIPE_ALL_CPU, &tracing_pipe_fops);
4288 trace_create_file("buffer_size_kb", 0644, d_tracer,
4289 &global_trace, &tracing_entries_fops);
4291 trace_create_file("trace_marker", 0220, d_tracer,
4292 NULL, &tracing_mark_fops);
4294 trace_create_file("saved_cmdlines", 0444, d_tracer,
4295 NULL, &tracing_saved_cmdlines_fops);
4297 trace_create_file("trace_clock", 0644, d_tracer, NULL,
4300 #ifdef CONFIG_DYNAMIC_FTRACE
4301 trace_create_file("dyn_ftrace_total_info", 0444, d_tracer,
4302 &ftrace_update_tot_cnt, &tracing_dyn_info_fops);
4304 #ifdef CONFIG_SYSPROF_TRACER
4305 init_tracer_sysprof_debugfs(d_tracer);
4308 create_trace_options_dir();
4310 for_each_tracing_cpu(cpu)
4311 tracing_init_debugfs_percpu(cpu);
4316 static int trace_panic_handler(struct notifier_block *this,
4317 unsigned long event, void *unused)
4319 if (ftrace_dump_on_oops)
4324 static struct notifier_block trace_panic_notifier = {
4325 .notifier_call = trace_panic_handler,
4327 .priority = 150 /* priority: INT_MAX >= x >= 0 */
4330 static int trace_die_handler(struct notifier_block *self,
4336 if (ftrace_dump_on_oops)
4345 static struct notifier_block trace_die_notifier = {
4346 .notifier_call = trace_die_handler,
4351 * printk is set to max of 1024, we really don't need it that big.
4352 * Nothing should be printing 1000 characters anyway.
4354 #define TRACE_MAX_PRINT 1000
4357 * Define here KERN_TRACE so that we have one place to modify
4358 * it if we decide to change what log level the ftrace dump
4361 #define KERN_TRACE KERN_EMERG
4364 trace_printk_seq(struct trace_seq *s)
4366 /* Probably should print a warning here. */
4370 /* should be zero ended, but we are paranoid. */
4371 s->buffer[s->len] = 0;
4373 printk(KERN_TRACE "%s", s->buffer);
4378 static void __ftrace_dump(bool disable_tracing)
4380 static arch_spinlock_t ftrace_dump_lock =
4381 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
4382 /* use static because iter can be a bit big for the stack */
4383 static struct trace_iterator iter;
4384 unsigned int old_userobj;
4385 static int dump_ran;
4386 unsigned long flags;
4390 local_irq_save(flags);
4391 arch_spin_lock(&ftrace_dump_lock);
4399 if (disable_tracing)
4402 for_each_tracing_cpu(cpu) {
4403 atomic_inc(&global_trace.data[cpu]->disabled);
4406 old_userobj = trace_flags & TRACE_ITER_SYM_USEROBJ;
4408 /* don't look at user memory in panic mode */
4409 trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
4411 printk(KERN_TRACE "Dumping ftrace buffer:\n");
4413 /* Simulate the iterator */
4414 iter.tr = &global_trace;
4415 iter.trace = current_trace;
4416 iter.cpu_file = TRACE_PIPE_ALL_CPU;
4419 * We need to stop all tracing on all CPUS to read the
4420 * the next buffer. This is a bit expensive, but is
4421 * not done often. We fill all what we can read,
4422 * and then release the locks again.
4425 while (!trace_empty(&iter)) {
4428 printk(KERN_TRACE "---------------------------------\n");
4432 /* reset all but tr, trace, and overruns */
4433 memset(&iter.seq, 0,
4434 sizeof(struct trace_iterator) -
4435 offsetof(struct trace_iterator, seq));
4436 iter.iter_flags |= TRACE_FILE_LAT_FMT;
4439 if (find_next_entry_inc(&iter) != NULL) {
4442 ret = print_trace_line(&iter);
4443 if (ret != TRACE_TYPE_NO_CONSUME)
4444 trace_consume(&iter);
4447 trace_printk_seq(&iter.seq);
4451 printk(KERN_TRACE " (ftrace buffer empty)\n");
4453 printk(KERN_TRACE "---------------------------------\n");
4455 /* Re-enable tracing if requested */
4456 if (!disable_tracing) {
4457 trace_flags |= old_userobj;
4459 for_each_tracing_cpu(cpu) {
4460 atomic_dec(&global_trace.data[cpu]->disabled);
4466 arch_spin_unlock(&ftrace_dump_lock);
4467 local_irq_restore(flags);
4470 /* By default: disable tracing after the dump */
4471 void ftrace_dump(void)
4473 __ftrace_dump(true);
4476 __init static int tracer_alloc_buffers(void)
4482 if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
4485 if (!alloc_cpumask_var(&tracing_cpumask, GFP_KERNEL))
4486 goto out_free_buffer_mask;
4488 /* To save memory, keep the ring buffer size to its minimum */
4489 if (ring_buffer_expanded)
4490 ring_buf_size = trace_buf_size;
4494 cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
4495 cpumask_copy(tracing_cpumask, cpu_all_mask);
4497 /* TODO: make the number of buffers hot pluggable with CPUS */
4498 global_trace.buffer = ring_buffer_alloc(ring_buf_size,
4499 TRACE_BUFFER_FLAGS);
4500 if (!global_trace.buffer) {
4501 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
4503 goto out_free_cpumask;
4505 global_trace.entries = ring_buffer_size(global_trace.buffer);
4508 #ifdef CONFIG_TRACER_MAX_TRACE
4509 max_tr.buffer = ring_buffer_alloc(ring_buf_size,
4510 TRACE_BUFFER_FLAGS);
4511 if (!max_tr.buffer) {
4512 printk(KERN_ERR "tracer: failed to allocate max ring buffer!\n");
4514 ring_buffer_free(global_trace.buffer);
4515 goto out_free_cpumask;
4517 max_tr.entries = ring_buffer_size(max_tr.buffer);
4518 WARN_ON(max_tr.entries != global_trace.entries);
4521 /* Allocate the first page for all buffers */
4522 for_each_tracing_cpu(i) {
4523 global_trace.data[i] = &per_cpu(global_trace_cpu, i);
4524 max_tr.data[i] = &per_cpu(max_tr_data, i);
4527 trace_init_cmdlines();
4529 register_tracer(&nop_trace);
4530 current_trace = &nop_trace;
4531 #ifdef CONFIG_BOOT_TRACER
4532 register_tracer(&boot_tracer);
4534 /* All seems OK, enable tracing */
4535 tracing_disabled = 0;
4537 atomic_notifier_chain_register(&panic_notifier_list,
4538 &trace_panic_notifier);
4540 register_die_notifier(&trace_die_notifier);
4545 free_cpumask_var(tracing_cpumask);
4546 out_free_buffer_mask:
4547 free_cpumask_var(tracing_buffer_mask);
4552 __init static int clear_boot_tracer(void)
4555 * The default tracer at boot buffer is an init section.
4556 * This function is called in lateinit. If we did not
4557 * find the boot tracer, then clear it out, to prevent
4558 * later registration from accessing the buffer that is
4559 * about to be freed.
4561 if (!default_bootup_tracer)
4564 printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
4565 default_bootup_tracer);
4566 default_bootup_tracer = NULL;
4571 early_initcall(tracer_alloc_buffers);
4572 fs_initcall(tracer_init_debugfs);
4573 late_initcall(clear_boot_tracer);